1  /* { dg-do compile } */
       2  /* { dg-options "-O2 -ftree-vectorize -march=armv8-a+sve --param aarch64-sve-compare-costs=0" } */
       3  
       4  #include <stdint.h>
       5  
       6  #define DEF_LOOP(TYPE)							\
       7    void __attribute__ ((noinline, noclone))				\
       8    test_##TYPE (int *restrict a, TYPE *restrict b, int a1, int a2,	\
       9  	       int a3, int a4, int n)					\
      10    {									\
      11      for (int i = 0; i < n; i += 2)					\
      12        {									\
      13  	a[i] = a[i] >= 1 & b[i] != 3 ? a1 : a2;				\
      14  	a[i + 1] = a[i + 1] >= 1 & b[i + 1] != 3 ? a3 : a4;		\
      15        }									\
      16    }
      17  
      18  #define FOR_EACH_TYPE(T) \
      19    T (int8_t) \
      20    T (uint8_t) \
      21    T (int16_t) \
      22    T (uint16_t) \
      23    T (int64_t) \
      24    T (uint64_t) \
      25    T (double)
      26  
      27  FOR_EACH_TYPE (DEF_LOOP)
      28  
      29  /* { dg-final { scan-assembler-times {\tld1b\t} 2 } } */
      30  /* { dg-final { scan-assembler-times {\tld1h\t} 2 } } */
      31  /* 4 for each 8-bit function, 2 for each 16-bit function, 1 for
      32     each 64-bit function.  */
      33  /* { dg-final { scan-assembler-times {\tld1w\t} 15 } } */
      34  /* 3 64-bit functions * 2 64-bit vectors per 32-bit vector.  */
      35  /* { dg-final { scan-assembler-times {\tld1d\t} 6 } } */
      36  /* { dg-final { scan-assembler-times {\tsel\tz[0-9]} 15 } } */